Detecting Anomalous Events in a Discriminator of an Embedded Device

2. The apparatus of claim 1, wherein the image comprises a spectrogram.

3. The apparatus of claim 1, wherein the trained discriminator is to determine whether the real world information comprises the positive condition comprising at least one of a glass break sound, a gunshot sound, a baby cry, a falling body, or an emergency siren.

4. The apparatus of claim 3, wherein the generator is to generate the synthetic samples including at least some synthetic glass break sounds or at least some synthetic gunshot sounds.

5. The apparatus of claim 1, further comprising a non-volatile storage comprising a plurality of binary classifiers, each associated with a given anomaly.

6. The apparatus of claim 1, wherein the discriminator is to output a probability metric comprising the indication of the anomaly.

7. The apparatus of claim 6, wherein the controller is to receive the probability metric, and in response to the probability metric exceeding a threshold, the controller is to trigger the action comprising a wireless communication of an alert to a remote location.

8. The apparatus of claim 7, further comprising a wireless circuit to send the wireless communication of the alert to the remote location and receive a command from the remote location, wherein in response to the command the apparatus is to perform the action.

9. The apparatus of claim 1, wherein the real world information comprises audio information and the signal processor is to process the audio information into a spectrogram.

10. The apparatus of claim 1, wherein the generative adversarial network comprises the discriminator and the generator, the apparatus not including the generator.

11. The apparatus of claim 10, wherein the generator and the discriminator are iteratively trained prior to provision of the discriminator to the apparatus.

13. The system of claim 12, wherein the discriminator comprises a binary classifier, the binary classifier to output a probability value, wherein when the probability value exceeds a threshold the anomaly is detected and when the probability value does not exceed the threshold no anomaly is detected.

14. The system of claim 12, further comprising a non-volatile memory to store the discriminator, wherein the system is to receive an updated discriminator for storage in the non-volatile memory as an in-field update.

16. The wireless device of claim 15, wherein the discriminator is to output a probability value, wherein when the probability value exceeds a threshold the anomaly is detected and when the probability value does not exceed the threshold no anomaly is detected.

17. The wireless device of claim 15, wherein the discriminator is trained with samples of a positive condition comprising the anomaly, the samples of the positive condition comprising the anomaly comprising synthetic samples generated by the generative adversarial network and real samples of the anomaly.